So as with a lot of matters in intelligence work it's subject to cost benefit calcs. If using it against a given target means they are incredibly unlikely to notice and it can then be used again and again, it doesn't take much target value for a government to deploy it which pushes towards more mass use. On the opposite end if using it means it will immediately become useless ever again, then the expected target value has to at least exceed the market cost (which itself will rise more quickly if 0-days are being consumed more quickly vs production), every time. In between is a spectrum of less or more use. Apple wants it as far towards "use it and lose it" as possible, but Trevor Perrin's argument makes sense here: even a relatively small increase in percentage of "use it and lose it" amongst the population could significantly change the mean weighted cost for threat actors.
If they could know for sure whether a given counter measure was deployed that'd reduce the cost again, but if they can't there is indeed a population benefit. It's like a mine field, there don't have to be that many mines scattered around to really hurt people's willingness to cross it!
